Thomas Cornell

A Type-Logical Perspective on Minimalist Derivations


Proceedings of the Formal Grammar Conference, Aix-en-Provence, August 1997.
12 pp., DVI (50kb); Postscript (150kb) 1-up; Postscript gzip-compressed (60kb) 1-up , 2-up.

The Handout

The talk presented at the Formal Grammar conference presented a rather more advanced and, frankly, more capable system than the one presented in the "precedings" paper. The handout for that talk is quite prosy and should be pretty readable on its own. In any case, enough people have expressed an interest that I thought I would include it here (postscript, gzip-compressed, 68kb).


Abstract

Many researchers in the past few years have remarked on the apparent relation between ideas from Chomsky's minimalist program and fundamental ideas in categorial grammar. The main feature of minimalist derivations which has suggested a connection with categorial grammar is the treatment of feature checking in Chapter 4 of Chomsky (1995). The idea that a feature on a moving phrase might be cancelled against a feature on the structure to which it moves is very reminiscent of the basic cancellation rules of categorial grammar. Seen in this light, the derivational need to cancel features reduces to the proof-theoretic need to reduce a hypothesis to axioms. This paper represents a first stab at the problem of developing ideas from the minimalist program within a type-logical deductive grammar framework. Some positive results indicate that the representational approach to minimalism of Brody (1995), involving crucially the idea of ``presyntactic chain formation'', can be formally related to a more strictly derivational approach by using a labeling algebra which simply reflects the phrase structure tree underlying the feature-driven derivation. A surprising negative result suggests that, given Chomsky's assumptions about the structure of the initial numeration, the generalized transformation Merge may be harder to implement in this setting than Move.


Last modified October 6, 1997
Tom Cornell's manuscript page,
email: cornell@sfs.nphil.uni-tuebingen.de